Firecrawl Scrape

Run firecrawl scrape --help for the full option list.

Done when: the page content or provider result has been checked for errors and inspected in bounded sections to answer the request. Preserve source links and disclose partial results.

Find tools, inspect inputs, and get help

Use the CLI help to check supported options rather than guessing:

Domain discovery with --domain-tools returns tool summaries by default. Add --tool-detail full for contracts upfront, or inspect one selected tool with list as shown below. Use --tool-detail compact for only provider, capability and description; inspect by those two IDs with list. Summary remains the default. Prefer full when several related contracts will be needed immediately.

For structured data, search for the task, inspect a matching tool’s contract, then execute with the exact input fields it declares:

Normal search includes web results and tool matches; search alexandria searches tools only. list <provider> <capability> --pretty shows the selected contract; use --json for machine-readable output. To browse progressively, use list, then list <category> --category, then list <provider>. Search and list do not execute the selected provider tool. Read only the contracts needed for the task; use returned identifiers rather than guessing them.

Read the expanded contract before building inputs or parsing results:

• required: true requires that input; each requiresOneOf group requires at least one member, not all of them.
• Selected-contract inspection already requests examples. Read the singular example.request and example.response when present; an empty request can be valid for tools with optional inputs.
• response.key identifies the records field inside data.alexandria[i].data; an empty key means that data object itself. Do not assume every provider returns records.
• Provider pagination differs from catalogue next: use the contract’s continuation input and the returned page/cursor, preserve filters, and stop at its exhaustion signal. paginated: true alone does not specify that mapping.

Execution and large results

URL scraping does not execute provider tools automatically. Use exact discovered input fields and resolve record IDs with lookup tools rather than inventing them. Check each data.alexandria[] result for errors, not just the outer success flag.

If the client reports an output/context limit, the upstream request may have succeeded. Preserve the request or scrape ID and recover the retained result before repeating the provider call. For large datasets and PDFs, save output with --json -o when a local filesystem is available and inspect bounded sections with jq or other file tools. Keep stderr separate from JSON stdout; do not merge streams with 2>&1 when piping to a JSON parser. Where remote processing is preferable, use firecrawl scrape firecrawl/bash to select from a retained result. Read large-result recovery (opens in a new tab) for IDs, command examples, expiry, and errors. This is explicit recovery, not automatic overflow detection.

PDFs and page budgets

PDFs cost 1 credit per parsed page. Use --max-pages (an integer from 1 to 10000) to limit PDF parsing, especially for large or unknown documents:

The cap applies to each PDF, not the whole command or total credits. Extra formats and options can add charges. The CLI does not quote page counts or costs before execution. Use JSON output to inspect the returned metadata.numPages (parsed), metadata.totalPages (document total), and metadata.creditsUsed when present; a smaller parsed count means the result is partial.

Tips

• Prefer plain scrape over --query. Scrape to a file, then use grep, head, or read the markdown directly — you can search and reason over the full content yourself. Use --query only when you want a single targeted answer without saving the page (costs 5 extra credits).
• Scrape handles static pages and JS-rendered SPAs. Escalate to interact when the page needs interaction (clicks, form fills, pagination) or scrape misses content.
• Multiple URLs are scraped concurrently — check firecrawl --status for your concurrency limit. This mode saves markdown only and ignores -o; other requested formats are dropped. If markdown wasn’t requested, the whole JSON response is written into the .md file.
• Single format outputs raw content. Multiple formats (e.g., --format markdown,links) output JSON.
• Always quote URLs — shell interprets ? and & as special characters.
• Naming convention: .firecrawl/{site}-{path}.md

See also

• firecrawl-search — find pages when you don’t have a URL
• firecrawl-interact — when scrape can’t get the content, use interact to click, fill forms, etc.
• firecrawl-download — bulk download an entire site to local files
• firecrawl-build-scrape (opens in a new tab) — building scrape into an app instead of running it here